We compare the performance of a recently proposed empirical climate model based on astronomical harmonics against all available general circulation climate models (GCM) used by the IPCC (2007) to interpret the 20th century global surface temperature. The proposed model assumes that the climate is resonating with, or synchronized to a set of natural harmonics that have been associated to the solar system planetary motion, mostly determined by Jupiter and Saturn. We show that the GCMs fail to reproduce the major decadal and multidecadal oscillations found in the global surface temperature record from 1850 to 2011. On the contrary, the proposed harmonic model is found to well reconstruct the observed climate oscillations from 1850 to 2011, and it is able to forecast the climate oscillations from 1950 to 2011 using the data covering the period 1850-1950, and vice versa. The 9.1-year cycle is shown to be likely related to a decadal Soli/Lunar tidal oscillation, while the 10-10.5, 20-21 and 60-62 year cycles are synchronous to solar and heliospheric planetary oscillations. Finally, we show how the presence of these large natural cycles can be used to correct the IPCC projected anthropogenic warming trend for the 21st century. By combining this corrected trend with the natural cycles, we show that the temperature may not significantly increase during the next 30 years mostly because of the negative phase of the 60-year cycle. The same IPCC projected anthropogenic emissions would imply a global warming by about 0.3-1.2 K by 2100, contrary to the IPCC 1.0-3.6 K projected warming. The results of this paper reinforce previous claims that the relevant physical mechanisms that explain the detected climatic cycles are still missing in the current GCMs and that climate variations at the multidecadal scales are astronomically induced and, in first approximation, can be forecast.